A Method to Extract Spots from the image of the ELISA (Enzyme-Linked Immunosorbent Assay) Spot Assay

A Method to Extract Spots from the image of the EISA (Enzyme-inked Immnosorbent Assa Spot Assay Chih-Yang in 1, Y-Tai Ching 1, B. A.W-Hsieh 1 Department of Compter and Information Science, National Chiao
of 5
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Related Documents
A Method to Extract Spots from the image of the EISA (Enzyme-inked Immnosorbent Assa Spot Assay Chih-Yang in 1, Y-Tai Ching 1, B. A.W-Hsieh 1 Department of Compter and Information Science, National Chiao Tng Uniersity, HsinCh, Taiwan Gradate Institte of Immnology, National Taiwan Uniersity College of Medicine,Taipei, Taiwan Abstract- EISA (Enzyme-inked Immnosorbent Assa Spot Assay is a method widely sed by immnologists to enmerate cytokine -prodcing cells within a specific cell poplation. The reslt of EISA is presented in an image containing nmeros color spots. We present a method to extract and to cont the nmber of spots. The proposed method is based on color analysis. Since CIE space hae linear perceptibility of color differences, we conert the RGB space to space. The system is trained to obtain the standard color of the spots and get the color difference image in space. According to the featre of the spots we design a special matched filters to filter ot the noise and enhance the spots. Finally a binary image is obtained. In the binary images, pixels in the spots hae gray scale 55 and the others are 0. Or design makes it easy to analyze the perimeter and size of the spots in addition to conting them in the binary image. Keywords: EISA, Spot, Segmentation, CIE. I. INTRODUCTION EISA Spot Assay is designed to detect cells that prodce cytokines [1]. Cytokines are proteins readily secreted by immne cells pon stimlation by antigens they recognize. Test wells are coated with anti-cytokine antibody before cells were added. A certain nmber of cells and antigen are added to the precoated wells. Dring incbation, cells are stimlated to secrete cytokine. The secreted cytokine is captred by the precoated antibody. After washing, an enzyme-conjgated secondary anti-cytokine antibody and the sbstrate are added in seqence. A color reaction (red in this case) specific to cytokine-secreting cells occrs as a reslt of enzymatic reaction. Each red spot represents one cytokine-secreting cell. Since there can be hndreds of spots in a well of 70 mm diameter, conting the spots is labor intensie een one ses a dissecting microscope. The goal of this work is to prodce a binary image containing pixels only in the spots. The first step in the presented method is color space conersion. It linearizes the perceptibility of color differences and proides a niform color space. The second step is to train the system by sing the color of the spots to obtain an image of color differentials. The third step is to apply the matched filter to identify the spots and remoe the ndesired noise. Finally, the reslted image in step three is thresholded according to the histogram and gets the binary image containing only pixels in spots. In the next section, we describe these steps. The reslts are shown in Section 3. II. METHOD 1. Color Space Conersion The spot colors proide good information for extracting the area of the spots. There were sed to be oer 40 color difference formlas before that CIE (Commission Internationale de I Eclairage) recommended two standard color difference formlas: the CIE a b and the CIE for srface and lighting indstries []. They attempted to linearize the perceptibility of color differences. They proided a niform color space. We hae chosen the space [3]. While ideo cameras se RGB representation for colors. We mst conert the representation from RGB to. The CIE also has recommended the se of other color-coordinate systems, deried from CIE XYZ. We conert the representation from RGB to XYZ space sing Eq. (1). X R Y = G. (1) Z B is based directly on CIE XYZ. The non-linear relations for,, and are gien below: Y 1/ 3 Y 116( ) 16 if Y n =, Y Y 903.3( ) 16 if ' ' = 13 ( n), (3) ' ' = 13 ( n ) ' 4X = ( X + 15Y + 3Z), (4) ' 9Y = ( X + 15Y + 3Z ) where n ' and n ' refer to the reference white X n, Y n and Z n. The daylight standard D 65 was sed as reference illminant. The non-linear relationship for Y is intended to mimic the logarithmic response of the eye. Now we hae conerted the spot image in RGB space denoted as f ( to space denoted as f (. RGB. Train the system and obtain an image of color differentials To train the system to recognize the standard color of spots, we se a ser interface method to select the spot area denoted as. We sed all pixels in the spot area to obtain the standard color denoted as ( µ, µ, µ ). We calclate the aerage for. Report Docmentation Page Report Date 5 Oct 001 Report Type N/A Dates Coered (from... to) - Title and Sbtitle A Method to Extract Spots From the Image of the EISA (Enzyme-inked Immnosorbent Assa Spot Assay Contract Nmber Grant Nmber Program Element Nmber Athor(s) Project Nmber Task Nmber Work Unit Nmber Performing Organization Name(s) and Address(es) Department of Compter and Information Science Natinal Chiao Tng Uniersity HsinCh, Taiwan Sponsoring/Monitoring Agency Name(s) and Address(es) US Army Research, Deelopment & Standardization Grop (UK) PSC 80 Box 15 FPO AE Performing Organization Report Nmber Sponsor/Monitor s Acronym(s) Sponsor/Monitor s Report Nmber(s) Distribtion/Aailability Statement Approed for pblic release, distribtion nlimited Spplementary Notes Papers from 3rd Annal International Conference of the IEEE Engineering in Medicine and Biology Society, Oct 5-8, 001, held in Istanbl, Trkey. See also ADM for entire conference on cd-rom., The original docment contains color images. Abstract Sbject Terms Report Classification Classification of Abstract Classification of this page imitation of Abstract UU Nmber of Pages 4 µ µ µ, (5) where N is the nmber of pixels in the spot area. The only ser interface reqired is to obtain the standard spot µ, µ, µ ) in this step. No more other ser color ( actiities are needed in the following steps. The difference between two measred colors can be expressed sing the CIE color difference formla. 0.5 E ( ) + ( ) + ( ) ). (6) f ( and ( µ, µ, µ ) to Eq. (6), We sbstitte then we can obtain the color difference image, f (, which is the difference between each pixel of f ( and the aerage spot pixel ( µ, µ, µ ). 3. Designing the Matched Filter The spots hae an approximate shape of circle. In fact, the intensity in a spot almost follows a -D Gassian fnction as Eq. (7) in the following x + y ) r g1 ( = e = e. (7) In Eq. (7) r is the radial distance measred from the center of the spot. If we define R as the radis at which the intensity drops to one half of its maximm ale, we can write the spot profile fnction as g We can simplify Eq. (8) as ( r / R)ln() ln( r / R ) = e = e. (8) x + y ) / R g =. (9) Now let s consider the detection of an arbitrary signal s(t) in additie Gassian white noise. If the signal is passed throgh a filter with transform fnction H(f), the otpt signal s O (t) is gien by ( j πft ) df so ( t) = H ( f ){ S ( f ) + η ( f )} e, (10) where S(f) is the Forier transform of s(, and η( f ) is the noise spectrm. By Schwartz s ineqality, it can be proed that the filter H(f), that maximizes the otpt SNR (Signal to Noise Ratio), is gien by Ho(f) =S (f) [4,5]. We apply inerse Forier transform for both Ho(f) and S (f). Since the inpt signal s(t) is a real ale, ho(t)=s(-t). Hence, in a -D case ho(=s(--. The optimal filter mst hae the same shape as the intensity profile. In other words, the optimal filter is gien by h( = s(. (11) We sbstitte Eq. (9) for Eq. (11) h (, (1) where g is a symmetry fnction, ths g. This optimal filter with the implse response h( is commonly known as the matched filter[6,7]. To obtain a zero mean filter, g is sbtracted by the mean of the filter. x + y )/ R g( m = m, (13) where m is the mean of the filter. The radis sizes of spots are ranged between 4 and 16 pixels. We only se pixel sizes 4, 8, 1 and 16 as the radis for designing the kernel of the matched filers. Using more matched filter kernels cold get more accrate reslts, bt also takes more compting time. Fig. 1 shows two of the 4 kernels of matched filters. After applying matched filter, Most of noises cold be remoed and ths the spots were enhanced. It is sefl for the following step. (a) (b) Fig. 1. -D matched filters sed in or proposed method. The radis size is (a) 4, and (b) Obtain the Binary Image Finally, we threshold the reslted images based on the histogram analysis to get binary images. Now we can easily to cont the nmber of spots, size of spots, and measre perimeter III. RESUTS In this section, seeral images were tested and the reslts obtained by the proposed method are presented. The inpt images are color images in Jpeg format of size 1600 by 100 pixels (Fig. -x(a), where x means 1,, and 3). Fig. -x(b) shows the distance measre. The reslts after matched filtering were applied are shown in Fig. -x(c). The reslts after threshold are shown in Fig. -x(d). The borders of spots Fig. -x(d) are oerlaid to the initial image are shown in Fig. -x(e). The proposed methods were implemented on a PC with a Pentim III (800 MHz) CPU rnning on Windows 000 operating system. The oerall exection time for a 1600x100 pixels image took less than 5 mintes. The radis sizes sed for filter kernel were 1,,3 and 4, respectiely. It took a qit long time for a 1600x100 image. Ths we redce the resoltion from 1600x100 to 800x600 and 400x300 for those images, respectiely. The 800x600 images took abot 5 seconds. The radis sizes sed for filter kernel are,4,6 and 8, respectiely. The exection time for 800x600 images is mch faster than the preios case. In the case of 400x300 images, it took abot only 3 seconds. The radis sizes sed for filter kernel are 1,,3 and 4, respectiely. The reslts of 1600x100, 800x600, and 400x300 hae almost the same qality, bt the exection time is significantly different. Becase the larger images the larger the radis sizes are needed for matched filter kernels. Since the operations of -D conoltion need large comptation time, we sggest sing 400x300 or 800x600 pixel images instead of 1600x (e) -1. (a) -. (a) -1. (b) -. (b) -1. (c) -. (c) -1. (d) -. (d) -. (e) -3 (e) Fig.. -x(a) are the tested images for proposed method. - x(b) are the color distance images of -x(a). -x(c) are the reslts after matched filters. -x(d) are the final reslts. -x(e) are the final reslts oerlying the border of -x(d) REFERENCES -3. (a) -3. (b) [1] K. Sorensen, and U. Brodbeck, Assessment of coatingefficiency in EISA plates by direct protein determination, Jorn. of Imm. Methods, 95: 91-93, [] Commission Internationale de I Eclairage Colorimetry, nd edn, CIE blication 15.. Paris: CIE, [3] G. Wyszecki, W.S. Stiles, Color Science: Concepts and Mthods, Qantitatie Data and Formlae, Second Ed. New York: Wiley, 198 [4] S. Haykin, Commnication Systems, New Delhi: Wiley Eastern, [5] D. H. Friedman, Detection of Signals by Template Matching, Baltimore, MD:Johns Hopkins Uniersity Press, [6] G.. Trin, An Introdction to Matched Filter, IRE Transcations in Information Theory, Jne [7] D. Middleton, On New Classes of Matched Filters and Generalizations of the matched filter concept, IRE Transactions on Information Theory, pp , Jne (c) -3. (d)
Similar documents
View more...
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks